Electrically-stimulated sensors for spectroscopy and tomography have been widely adopted because they are non-invasive, relatively inexpensive, and straightforward in implementation. Despite decades of development, their widespread adoption is limited partly due to their low-spatial resolution. In this work, we propose a technique based on the electrical impedance tomography sensing principle without the image reconstruction and tomographic processing modules. By tailoring the stimulation and measurement protocols and evaluating the raw data output for markers of highly discernible patterns, we demonstrate the technique's ability in identifying minute object variance in key physical parameters such as movement, size, shape, and conductivity. With a bare-bones hardware system setup, we observed that raw impedance data are discernably sensitive to minute variations within a typical electrical impedance phantom. With generic machine learning models, we further reveal these signal patterns are autonomously classifiable at high accuracy, leading to a sensitive, and operationally simplistic sensing approach adaptable to applications such as biophysiological sensing.